Chemokine Receptors-Structure-Based Virtual Screening Assisted by Machine Learning.

Area of Science:

• Immunology and Pharmacology
• Computational Chemistry
• Drug Discovery

Background:

• Chemokines regulate immune cell migration and are implicated in inflammatory diseases.
• Conventional chemokine receptors (CCR1-6) are G protein-coupled receptors linked to multiple sclerosis, asthma, and rheumatoid arthritis.
• Structure-based drug design requires detailed receptor information.

Purpose of the Study:

• To evaluate the utility of available chemokine receptor structures for drug design.
• To identify novel active compounds for CCR2 and CCR3 using structure-based virtual screening.
• To develop and compare machine learning models for ligand-based drug design of CCR inhibitors.

Main Methods:

• Collected crystal, cryo-EM, and homology models for CCR1-6.
• Performed structure-based virtual screening targeting CCR2 and CCR3.
• Utilized known CCR inhibitors from ChEMBL as training data for machine learning models (LightGBM and Keras/TensorFlow NN).
• Assessed the predictive performance of virtual screening and machine learning methods for CCR2/CCR3 activity and selectivity.

Main Results:

• Structure-based virtual screening identified several novel active compounds for CCR2 and CCR3.
• Machine learning models were trained using known CCR inhibitors.
• A combination of methods successfully proposed active ligands for CCR2 and CCR3.
• Two compounds were consistently predicted as CCR3 active by Glide, Keras/TensorFlow NN, and LightGBM.
• The study assessed the subtype selectivity prediction capabilities of the tested methods.

Conclusions:

• Available chemokine receptor structures are valuable for structure-based drug design.
• Integrated structure-based virtual screening and machine learning approaches can effectively identify potent chemokine receptor modulators.
• This study provides promising lead compounds for CCR2 and CCR3 and advances computational methods for drug discovery in inflammatory diseases.